Anru Zhang - Cross: efficient low-rank tensor completion
Slide 26: very impressive statement: With only around `0.3% observable entries, we can approximate the low-rank structure' as well as if we observe all entries.
By what measure are you saying `as good/well as' observing all entries? How are you measuring the performance relative to the other case?
To take away from your talk: why do you think your algorithm does so well?